KVM: add MSR based hypercall API
[powerpc.git] / drivers / kvm / svm.c
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * AMD SVM support
5  *
6  * Copyright (C) 2006 Qumranet, Inc.
7  *
8  * Authors:
9  *   Yaniv Kamay  <yaniv@qumranet.com>
10  *   Avi Kivity   <avi@qumranet.com>
11  *
12  * This work is licensed under the terms of the GNU GPL, version 2.  See
13  * the COPYING file in the top-level directory.
14  *
15  */
16
17 #include <linux/module.h>
18 #include <linux/kernel.h>
19 #include <linux/vmalloc.h>
20 #include <linux/highmem.h>
21 #include <linux/profile.h>
22 #include <asm/desc.h>
23
24 #include "kvm_svm.h"
25 #include "x86_emulate.h"
26
27 MODULE_AUTHOR("Qumranet");
28 MODULE_LICENSE("GPL");
29
30 #define IOPM_ALLOC_ORDER 2
31 #define MSRPM_ALLOC_ORDER 1
32
33 #define DB_VECTOR 1
34 #define UD_VECTOR 6
35 #define GP_VECTOR 13
36
37 #define DR7_GD_MASK (1 << 13)
38 #define DR6_BD_MASK (1 << 13)
39 #define CR4_DE_MASK (1UL << 3)
40
41 #define SEG_TYPE_LDT 2
42 #define SEG_TYPE_BUSY_TSS16 3
43
44 #define KVM_EFER_LMA (1 << 10)
45 #define KVM_EFER_LME (1 << 8)
46
47 unsigned long iopm_base;
48 unsigned long msrpm_base;
49
50 struct kvm_ldttss_desc {
51         u16 limit0;
52         u16 base0;
53         unsigned base1 : 8, type : 5, dpl : 2, p : 1;
54         unsigned limit1 : 4, zero0 : 3, g : 1, base2 : 8;
55         u32 base3;
56         u32 zero1;
57 } __attribute__((packed));
58
59 struct svm_cpu_data {
60         int cpu;
61
62         uint64_t asid_generation;
63         uint32_t max_asid;
64         uint32_t next_asid;
65         struct kvm_ldttss_desc *tss_desc;
66
67         struct page *save_area;
68 };
69
70 static DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
71
72 struct svm_init_data {
73         int cpu;
74         int r;
75 };
76
77 static u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
78
79 #define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
80 #define MSRS_RANGE_SIZE 2048
81 #define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
82
83 #define MAX_INST_SIZE 15
84
85 static unsigned get_addr_size(struct kvm_vcpu *vcpu)
86 {
87         struct vmcb_save_area *sa = &vcpu->svm->vmcb->save;
88         u16 cs_attrib;
89
90         if (!(sa->cr0 & CR0_PE_MASK) || (sa->rflags & X86_EFLAGS_VM))
91                 return 2;
92
93         cs_attrib = sa->cs.attrib;
94
95         return (cs_attrib & SVM_SELECTOR_L_MASK) ? 8 :
96                                 (cs_attrib & SVM_SELECTOR_DB_MASK) ? 4 : 2;
97 }
98
99 static inline u8 pop_irq(struct kvm_vcpu *vcpu)
100 {
101         int word_index = __ffs(vcpu->irq_summary);
102         int bit_index = __ffs(vcpu->irq_pending[word_index]);
103         int irq = word_index * BITS_PER_LONG + bit_index;
104
105         clear_bit(bit_index, &vcpu->irq_pending[word_index]);
106         if (!vcpu->irq_pending[word_index])
107                 clear_bit(word_index, &vcpu->irq_summary);
108         return irq;
109 }
110
111 static inline void push_irq(struct kvm_vcpu *vcpu, u8 irq)
112 {
113         set_bit(irq, vcpu->irq_pending);
114         set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
115 }
116
117 static inline void clgi(void)
118 {
119         asm volatile (SVM_CLGI);
120 }
121
122 static inline void stgi(void)
123 {
124         asm volatile (SVM_STGI);
125 }
126
127 static inline void invlpga(unsigned long addr, u32 asid)
128 {
129         asm volatile (SVM_INVLPGA :: "a"(addr), "c"(asid));
130 }
131
132 static inline unsigned long kvm_read_cr2(void)
133 {
134         unsigned long cr2;
135
136         asm volatile ("mov %%cr2, %0" : "=r" (cr2));
137         return cr2;
138 }
139
140 static inline void kvm_write_cr2(unsigned long val)
141 {
142         asm volatile ("mov %0, %%cr2" :: "r" (val));
143 }
144
145 static inline unsigned long read_dr6(void)
146 {
147         unsigned long dr6;
148
149         asm volatile ("mov %%dr6, %0" : "=r" (dr6));
150         return dr6;
151 }
152
153 static inline void write_dr6(unsigned long val)
154 {
155         asm volatile ("mov %0, %%dr6" :: "r" (val));
156 }
157
158 static inline unsigned long read_dr7(void)
159 {
160         unsigned long dr7;
161
162         asm volatile ("mov %%dr7, %0" : "=r" (dr7));
163         return dr7;
164 }
165
166 static inline void write_dr7(unsigned long val)
167 {
168         asm volatile ("mov %0, %%dr7" :: "r" (val));
169 }
170
171 static inline void force_new_asid(struct kvm_vcpu *vcpu)
172 {
173         vcpu->svm->asid_generation--;
174 }
175
176 static inline void flush_guest_tlb(struct kvm_vcpu *vcpu)
177 {
178         force_new_asid(vcpu);
179 }
180
181 static void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
182 {
183         if (!(efer & KVM_EFER_LMA))
184                 efer &= ~KVM_EFER_LME;
185
186         vcpu->svm->vmcb->save.efer = efer | MSR_EFER_SVME_MASK;
187         vcpu->shadow_efer = efer;
188 }
189
190 static void svm_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
191 {
192         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
193                                                 SVM_EVTINJ_VALID_ERR |
194                                                 SVM_EVTINJ_TYPE_EXEPT |
195                                                 GP_VECTOR;
196         vcpu->svm->vmcb->control.event_inj_err = error_code;
197 }
198
199 static void inject_ud(struct kvm_vcpu *vcpu)
200 {
201         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
202                                                 SVM_EVTINJ_TYPE_EXEPT |
203                                                 UD_VECTOR;
204 }
205
206 static void inject_db(struct kvm_vcpu *vcpu)
207 {
208         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
209                                                 SVM_EVTINJ_TYPE_EXEPT |
210                                                 DB_VECTOR;
211 }
212
213 static int is_page_fault(uint32_t info)
214 {
215         info &= SVM_EVTINJ_VEC_MASK | SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
216         return info == (PF_VECTOR | SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_EXEPT);
217 }
218
219 static int is_external_interrupt(u32 info)
220 {
221         info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
222         return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
223 }
224
225 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
226 {
227         if (!vcpu->svm->next_rip) {
228                 printk(KERN_DEBUG "%s: NOP\n", __FUNCTION__);
229                 return;
230         }
231         if (vcpu->svm->next_rip - vcpu->svm->vmcb->save.rip > 15) {
232                 printk(KERN_ERR "%s: ip 0x%llx next 0x%llx\n",
233                        __FUNCTION__,
234                        vcpu->svm->vmcb->save.rip,
235                        vcpu->svm->next_rip);
236         }
237
238         vcpu->rip = vcpu->svm->vmcb->save.rip = vcpu->svm->next_rip;
239         vcpu->svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
240
241         vcpu->interrupt_window_open = 1;
242 }
243
244 static int has_svm(void)
245 {
246         uint32_t eax, ebx, ecx, edx;
247
248         if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD) {
249                 printk(KERN_INFO "has_svm: not amd\n");
250                 return 0;
251         }
252
253         cpuid(0x80000000, &eax, &ebx, &ecx, &edx);
254         if (eax < SVM_CPUID_FUNC) {
255                 printk(KERN_INFO "has_svm: can't execute cpuid_8000000a\n");
256                 return 0;
257         }
258
259         cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
260         if (!(ecx & (1 << SVM_CPUID_FEATURE_SHIFT))) {
261                 printk(KERN_DEBUG "has_svm: svm not available\n");
262                 return 0;
263         }
264         return 1;
265 }
266
267 static void svm_hardware_disable(void *garbage)
268 {
269         struct svm_cpu_data *svm_data
270                 = per_cpu(svm_data, raw_smp_processor_id());
271
272         if (svm_data) {
273                 uint64_t efer;
274
275                 wrmsrl(MSR_VM_HSAVE_PA, 0);
276                 rdmsrl(MSR_EFER, efer);
277                 wrmsrl(MSR_EFER, efer & ~MSR_EFER_SVME_MASK);
278                 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
279                 __free_page(svm_data->save_area);
280                 kfree(svm_data);
281         }
282 }
283
284 static void svm_hardware_enable(void *garbage)
285 {
286
287         struct svm_cpu_data *svm_data;
288         uint64_t efer;
289 #ifdef CONFIG_X86_64
290         struct desc_ptr gdt_descr;
291 #else
292         struct Xgt_desc_struct gdt_descr;
293 #endif
294         struct desc_struct *gdt;
295         int me = raw_smp_processor_id();
296
297         if (!has_svm()) {
298                 printk(KERN_ERR "svm_cpu_init: err EOPNOTSUPP on %d\n", me);
299                 return;
300         }
301         svm_data = per_cpu(svm_data, me);
302
303         if (!svm_data) {
304                 printk(KERN_ERR "svm_cpu_init: svm_data is NULL on %d\n",
305                        me);
306                 return;
307         }
308
309         svm_data->asid_generation = 1;
310         svm_data->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
311         svm_data->next_asid = svm_data->max_asid + 1;
312
313         asm volatile ( "sgdt %0" : "=m"(gdt_descr) );
314         gdt = (struct desc_struct *)gdt_descr.address;
315         svm_data->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
316
317         rdmsrl(MSR_EFER, efer);
318         wrmsrl(MSR_EFER, efer | MSR_EFER_SVME_MASK);
319
320         wrmsrl(MSR_VM_HSAVE_PA,
321                page_to_pfn(svm_data->save_area) << PAGE_SHIFT);
322 }
323
324 static int svm_cpu_init(int cpu)
325 {
326         struct svm_cpu_data *svm_data;
327         int r;
328
329         svm_data = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
330         if (!svm_data)
331                 return -ENOMEM;
332         svm_data->cpu = cpu;
333         svm_data->save_area = alloc_page(GFP_KERNEL);
334         r = -ENOMEM;
335         if (!svm_data->save_area)
336                 goto err_1;
337
338         per_cpu(svm_data, cpu) = svm_data;
339
340         return 0;
341
342 err_1:
343         kfree(svm_data);
344         return r;
345
346 }
347
348 static int set_msr_interception(u32 *msrpm, unsigned msr,
349                                 int read, int write)
350 {
351         int i;
352
353         for (i = 0; i < NUM_MSR_MAPS; i++) {
354                 if (msr >= msrpm_ranges[i] &&
355                     msr < msrpm_ranges[i] + MSRS_IN_RANGE) {
356                         u32 msr_offset = (i * MSRS_IN_RANGE + msr -
357                                           msrpm_ranges[i]) * 2;
358
359                         u32 *base = msrpm + (msr_offset / 32);
360                         u32 msr_shift = msr_offset % 32;
361                         u32 mask = ((write) ? 0 : 2) | ((read) ? 0 : 1);
362                         *base = (*base & ~(0x3 << msr_shift)) |
363                                 (mask << msr_shift);
364                         return 1;
365                 }
366         }
367         printk(KERN_DEBUG "%s: not found 0x%x\n", __FUNCTION__, msr);
368         return 0;
369 }
370
371 static __init int svm_hardware_setup(void)
372 {
373         int cpu;
374         struct page *iopm_pages;
375         struct page *msrpm_pages;
376         void *msrpm_va;
377         int r;
378
379         kvm_emulator_want_group7_invlpg();
380
381         iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
382
383         if (!iopm_pages)
384                 return -ENOMEM;
385         memset(page_address(iopm_pages), 0xff,
386                                         PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
387         iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
388
389
390         msrpm_pages = alloc_pages(GFP_KERNEL, MSRPM_ALLOC_ORDER);
391
392         r = -ENOMEM;
393         if (!msrpm_pages)
394                 goto err_1;
395
396         msrpm_va = page_address(msrpm_pages);
397         memset(msrpm_va, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
398         msrpm_base = page_to_pfn(msrpm_pages) << PAGE_SHIFT;
399
400 #ifdef CONFIG_X86_64
401         set_msr_interception(msrpm_va, MSR_GS_BASE, 1, 1);
402         set_msr_interception(msrpm_va, MSR_FS_BASE, 1, 1);
403         set_msr_interception(msrpm_va, MSR_KERNEL_GS_BASE, 1, 1);
404         set_msr_interception(msrpm_va, MSR_LSTAR, 1, 1);
405         set_msr_interception(msrpm_va, MSR_CSTAR, 1, 1);
406         set_msr_interception(msrpm_va, MSR_SYSCALL_MASK, 1, 1);
407 #endif
408         set_msr_interception(msrpm_va, MSR_K6_STAR, 1, 1);
409         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_CS, 1, 1);
410         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_ESP, 1, 1);
411         set_msr_interception(msrpm_va, MSR_IA32_SYSENTER_EIP, 1, 1);
412
413         for_each_online_cpu(cpu) {
414                 r = svm_cpu_init(cpu);
415                 if (r)
416                         goto err_2;
417         }
418         return 0;
419
420 err_2:
421         __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
422         msrpm_base = 0;
423 err_1:
424         __free_pages(iopm_pages, IOPM_ALLOC_ORDER);
425         iopm_base = 0;
426         return r;
427 }
428
429 static __exit void svm_hardware_unsetup(void)
430 {
431         __free_pages(pfn_to_page(msrpm_base >> PAGE_SHIFT), MSRPM_ALLOC_ORDER);
432         __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
433         iopm_base = msrpm_base = 0;
434 }
435
436 static void init_seg(struct vmcb_seg *seg)
437 {
438         seg->selector = 0;
439         seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
440                 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
441         seg->limit = 0xffff;
442         seg->base = 0;
443 }
444
445 static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
446 {
447         seg->selector = 0;
448         seg->attrib = SVM_SELECTOR_P_MASK | type;
449         seg->limit = 0xffff;
450         seg->base = 0;
451 }
452
453 static int svm_vcpu_setup(struct kvm_vcpu *vcpu)
454 {
455         return 0;
456 }
457
458 static void init_vmcb(struct vmcb *vmcb)
459 {
460         struct vmcb_control_area *control = &vmcb->control;
461         struct vmcb_save_area *save = &vmcb->save;
462         u64 tsc;
463
464         control->intercept_cr_read =    INTERCEPT_CR0_MASK |
465                                         INTERCEPT_CR3_MASK |
466                                         INTERCEPT_CR4_MASK;
467
468         control->intercept_cr_write =   INTERCEPT_CR0_MASK |
469                                         INTERCEPT_CR3_MASK |
470                                         INTERCEPT_CR4_MASK;
471
472         control->intercept_dr_read =    INTERCEPT_DR0_MASK |
473                                         INTERCEPT_DR1_MASK |
474                                         INTERCEPT_DR2_MASK |
475                                         INTERCEPT_DR3_MASK;
476
477         control->intercept_dr_write =   INTERCEPT_DR0_MASK |
478                                         INTERCEPT_DR1_MASK |
479                                         INTERCEPT_DR2_MASK |
480                                         INTERCEPT_DR3_MASK |
481                                         INTERCEPT_DR5_MASK |
482                                         INTERCEPT_DR7_MASK;
483
484         control->intercept_exceptions = 1 << PF_VECTOR;
485
486
487         control->intercept =    (1ULL << INTERCEPT_INTR) |
488                                 (1ULL << INTERCEPT_NMI) |
489                 /*
490                  * selective cr0 intercept bug?
491                  *      0:   0f 22 d8                mov    %eax,%cr3
492                  *      3:   0f 20 c0                mov    %cr0,%eax
493                  *      6:   0d 00 00 00 80          or     $0x80000000,%eax
494                  *      b:   0f 22 c0                mov    %eax,%cr0
495                  * set cr3 ->interception
496                  * get cr0 ->interception
497                  * set cr0 -> no interception
498                  */
499                 /*              (1ULL << INTERCEPT_SELECTIVE_CR0) | */
500                                 (1ULL << INTERCEPT_CPUID) |
501                                 (1ULL << INTERCEPT_HLT) |
502                                 (1ULL << INTERCEPT_INVLPGA) |
503                                 (1ULL << INTERCEPT_IOIO_PROT) |
504                                 (1ULL << INTERCEPT_MSR_PROT) |
505                                 (1ULL << INTERCEPT_TASK_SWITCH) |
506                                 (1ULL << INTERCEPT_SHUTDOWN) |
507                                 (1ULL << INTERCEPT_VMRUN) |
508                                 (1ULL << INTERCEPT_VMMCALL) |
509                                 (1ULL << INTERCEPT_VMLOAD) |
510                                 (1ULL << INTERCEPT_VMSAVE) |
511                                 (1ULL << INTERCEPT_STGI) |
512                                 (1ULL << INTERCEPT_CLGI) |
513                                 (1ULL << INTERCEPT_SKINIT);
514
515         control->iopm_base_pa = iopm_base;
516         control->msrpm_base_pa = msrpm_base;
517         rdtscll(tsc);
518         control->tsc_offset = -tsc;
519         control->int_ctl = V_INTR_MASKING_MASK;
520
521         init_seg(&save->es);
522         init_seg(&save->ss);
523         init_seg(&save->ds);
524         init_seg(&save->fs);
525         init_seg(&save->gs);
526
527         save->cs.selector = 0xf000;
528         /* Executable/Readable Code Segment */
529         save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
530                 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
531         save->cs.limit = 0xffff;
532         /*
533          * cs.base should really be 0xffff0000, but vmx can't handle that, so
534          * be consistent with it.
535          *
536          * Replace when we have real mode working for vmx.
537          */
538         save->cs.base = 0xf0000;
539
540         save->gdtr.limit = 0xffff;
541         save->idtr.limit = 0xffff;
542
543         init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
544         init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
545
546         save->efer = MSR_EFER_SVME_MASK;
547
548         save->dr6 = 0xffff0ff0;
549         save->dr7 = 0x400;
550         save->rflags = 2;
551         save->rip = 0x0000fff0;
552
553         /*
554          * cr0 val on cpu init should be 0x60000010, we enable cpu
555          * cache by default. the orderly way is to enable cache in bios.
556          */
557         save->cr0 = 0x00000010 | CR0_PG_MASK;
558         save->cr4 = CR4_PAE_MASK;
559         /* rdx = ?? */
560 }
561
562 static int svm_create_vcpu(struct kvm_vcpu *vcpu)
563 {
564         struct page *page;
565         int r;
566
567         r = -ENOMEM;
568         vcpu->svm = kzalloc(sizeof *vcpu->svm, GFP_KERNEL);
569         if (!vcpu->svm)
570                 goto out1;
571         page = alloc_page(GFP_KERNEL);
572         if (!page)
573                 goto out2;
574
575         vcpu->svm->vmcb = page_address(page);
576         memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
577         vcpu->svm->vmcb_pa = page_to_pfn(page) << PAGE_SHIFT;
578         vcpu->svm->cr0 = 0x00000010;
579         vcpu->svm->asid_generation = 0;
580         memset(vcpu->svm->db_regs, 0, sizeof(vcpu->svm->db_regs));
581         init_vmcb(vcpu->svm->vmcb);
582
583         fx_init(vcpu);
584
585         return 0;
586
587 out2:
588         kfree(vcpu->svm);
589 out1:
590         return r;
591 }
592
593 static void svm_free_vcpu(struct kvm_vcpu *vcpu)
594 {
595         if (!vcpu->svm)
596                 return;
597         if (vcpu->svm->vmcb)
598                 __free_page(pfn_to_page(vcpu->svm->vmcb_pa >> PAGE_SHIFT));
599         kfree(vcpu->svm);
600 }
601
602 static struct kvm_vcpu *svm_vcpu_load(struct kvm_vcpu *vcpu)
603 {
604         get_cpu();
605         return vcpu;
606 }
607
608 static void svm_vcpu_put(struct kvm_vcpu *vcpu)
609 {
610         put_cpu();
611 }
612
613 static void svm_vcpu_decache(struct kvm_vcpu *vcpu)
614 {
615 }
616
617 static void svm_cache_regs(struct kvm_vcpu *vcpu)
618 {
619         vcpu->regs[VCPU_REGS_RAX] = vcpu->svm->vmcb->save.rax;
620         vcpu->regs[VCPU_REGS_RSP] = vcpu->svm->vmcb->save.rsp;
621         vcpu->rip = vcpu->svm->vmcb->save.rip;
622 }
623
624 static void svm_decache_regs(struct kvm_vcpu *vcpu)
625 {
626         vcpu->svm->vmcb->save.rax = vcpu->regs[VCPU_REGS_RAX];
627         vcpu->svm->vmcb->save.rsp = vcpu->regs[VCPU_REGS_RSP];
628         vcpu->svm->vmcb->save.rip = vcpu->rip;
629 }
630
631 static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
632 {
633         return vcpu->svm->vmcb->save.rflags;
634 }
635
636 static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
637 {
638         vcpu->svm->vmcb->save.rflags = rflags;
639 }
640
641 static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
642 {
643         struct vmcb_save_area *save = &vcpu->svm->vmcb->save;
644
645         switch (seg) {
646         case VCPU_SREG_CS: return &save->cs;
647         case VCPU_SREG_DS: return &save->ds;
648         case VCPU_SREG_ES: return &save->es;
649         case VCPU_SREG_FS: return &save->fs;
650         case VCPU_SREG_GS: return &save->gs;
651         case VCPU_SREG_SS: return &save->ss;
652         case VCPU_SREG_TR: return &save->tr;
653         case VCPU_SREG_LDTR: return &save->ldtr;
654         }
655         BUG();
656         return NULL;
657 }
658
659 static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
660 {
661         struct vmcb_seg *s = svm_seg(vcpu, seg);
662
663         return s->base;
664 }
665
666 static void svm_get_segment(struct kvm_vcpu *vcpu,
667                             struct kvm_segment *var, int seg)
668 {
669         struct vmcb_seg *s = svm_seg(vcpu, seg);
670
671         var->base = s->base;
672         var->limit = s->limit;
673         var->selector = s->selector;
674         var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
675         var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
676         var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
677         var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
678         var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
679         var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
680         var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
681         var->g = (s->attrib >> SVM_SELECTOR_G_SHIFT) & 1;
682         var->unusable = !var->present;
683 }
684
685 static void svm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
686 {
687         struct vmcb_seg *s = svm_seg(vcpu, VCPU_SREG_CS);
688
689         *db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
690         *l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
691 }
692
693 static void svm_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
694 {
695         dt->limit = vcpu->svm->vmcb->save.idtr.limit;
696         dt->base = vcpu->svm->vmcb->save.idtr.base;
697 }
698
699 static void svm_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
700 {
701         vcpu->svm->vmcb->save.idtr.limit = dt->limit;
702         vcpu->svm->vmcb->save.idtr.base = dt->base ;
703 }
704
705 static void svm_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
706 {
707         dt->limit = vcpu->svm->vmcb->save.gdtr.limit;
708         dt->base = vcpu->svm->vmcb->save.gdtr.base;
709 }
710
711 static void svm_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
712 {
713         vcpu->svm->vmcb->save.gdtr.limit = dt->limit;
714         vcpu->svm->vmcb->save.gdtr.base = dt->base ;
715 }
716
717 static void svm_decache_cr0_cr4_guest_bits(struct kvm_vcpu *vcpu)
718 {
719 }
720
721 static void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
722 {
723 #ifdef CONFIG_X86_64
724         if (vcpu->shadow_efer & KVM_EFER_LME) {
725                 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK)) {
726                         vcpu->shadow_efer |= KVM_EFER_LMA;
727                         vcpu->svm->vmcb->save.efer |= KVM_EFER_LMA | KVM_EFER_LME;
728                 }
729
730                 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK) ) {
731                         vcpu->shadow_efer &= ~KVM_EFER_LMA;
732                         vcpu->svm->vmcb->save.efer &= ~(KVM_EFER_LMA | KVM_EFER_LME);
733                 }
734         }
735 #endif
736         vcpu->svm->cr0 = cr0;
737         vcpu->svm->vmcb->save.cr0 = cr0 | CR0_PG_MASK | CR0_WP_MASK;
738         vcpu->cr0 = cr0;
739 }
740
741 static void svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
742 {
743        vcpu->cr4 = cr4;
744        vcpu->svm->vmcb->save.cr4 = cr4 | CR4_PAE_MASK;
745 }
746
747 static void svm_set_segment(struct kvm_vcpu *vcpu,
748                             struct kvm_segment *var, int seg)
749 {
750         struct vmcb_seg *s = svm_seg(vcpu, seg);
751
752         s->base = var->base;
753         s->limit = var->limit;
754         s->selector = var->selector;
755         if (var->unusable)
756                 s->attrib = 0;
757         else {
758                 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
759                 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
760                 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
761                 s->attrib |= (var->present & 1) << SVM_SELECTOR_P_SHIFT;
762                 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
763                 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
764                 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
765                 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
766         }
767         if (seg == VCPU_SREG_CS)
768                 vcpu->svm->vmcb->save.cpl
769                         = (vcpu->svm->vmcb->save.cs.attrib
770                            >> SVM_SELECTOR_DPL_SHIFT) & 3;
771
772 }
773
774 /* FIXME:
775
776         vcpu->svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
777         vcpu->svm->vmcb->control.int_ctl |= (sregs->cr8 & V_TPR_MASK);
778
779 */
780
781 static int svm_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
782 {
783         return -EOPNOTSUPP;
784 }
785
786 static void load_host_msrs(struct kvm_vcpu *vcpu)
787 {
788         int i;
789
790         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
791                 wrmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
792 }
793
794 static void save_host_msrs(struct kvm_vcpu *vcpu)
795 {
796         int i;
797
798         for ( i = 0; i < NR_HOST_SAVE_MSRS; i++)
799                 rdmsrl(host_save_msrs[i], vcpu->svm->host_msrs[i]);
800 }
801
802 static void new_asid(struct kvm_vcpu *vcpu, struct svm_cpu_data *svm_data)
803 {
804         if (svm_data->next_asid > svm_data->max_asid) {
805                 ++svm_data->asid_generation;
806                 svm_data->next_asid = 1;
807                 vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
808         }
809
810         vcpu->cpu = svm_data->cpu;
811         vcpu->svm->asid_generation = svm_data->asid_generation;
812         vcpu->svm->vmcb->control.asid = svm_data->next_asid++;
813 }
814
815 static void svm_invlpg(struct kvm_vcpu *vcpu, gva_t address)
816 {
817         invlpga(address, vcpu->svm->vmcb->control.asid); // is needed?
818 }
819
820 static unsigned long svm_get_dr(struct kvm_vcpu *vcpu, int dr)
821 {
822         return vcpu->svm->db_regs[dr];
823 }
824
825 static void svm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long value,
826                        int *exception)
827 {
828         *exception = 0;
829
830         if (vcpu->svm->vmcb->save.dr7 & DR7_GD_MASK) {
831                 vcpu->svm->vmcb->save.dr7 &= ~DR7_GD_MASK;
832                 vcpu->svm->vmcb->save.dr6 |= DR6_BD_MASK;
833                 *exception = DB_VECTOR;
834                 return;
835         }
836
837         switch (dr) {
838         case 0 ... 3:
839                 vcpu->svm->db_regs[dr] = value;
840                 return;
841         case 4 ... 5:
842                 if (vcpu->cr4 & CR4_DE_MASK) {
843                         *exception = UD_VECTOR;
844                         return;
845                 }
846         case 7: {
847                 if (value & ~((1ULL << 32) - 1)) {
848                         *exception = GP_VECTOR;
849                         return;
850                 }
851                 vcpu->svm->vmcb->save.dr7 = value;
852                 return;
853         }
854         default:
855                 printk(KERN_DEBUG "%s: unexpected dr %u\n",
856                        __FUNCTION__, dr);
857                 *exception = UD_VECTOR;
858                 return;
859         }
860 }
861
862 static int pf_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
863 {
864         u32 exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
865         u64 fault_address;
866         u32 error_code;
867         enum emulation_result er;
868         int r;
869
870         if (is_external_interrupt(exit_int_info))
871                 push_irq(vcpu, exit_int_info & SVM_EVTINJ_VEC_MASK);
872
873         spin_lock(&vcpu->kvm->lock);
874
875         fault_address  = vcpu->svm->vmcb->control.exit_info_2;
876         error_code = vcpu->svm->vmcb->control.exit_info_1;
877         r = kvm_mmu_page_fault(vcpu, fault_address, error_code);
878         if (r < 0) {
879                 spin_unlock(&vcpu->kvm->lock);
880                 return r;
881         }
882         if (!r) {
883                 spin_unlock(&vcpu->kvm->lock);
884                 return 1;
885         }
886         er = emulate_instruction(vcpu, kvm_run, fault_address, error_code);
887         spin_unlock(&vcpu->kvm->lock);
888
889         switch (er) {
890         case EMULATE_DONE:
891                 return 1;
892         case EMULATE_DO_MMIO:
893                 ++kvm_stat.mmio_exits;
894                 kvm_run->exit_reason = KVM_EXIT_MMIO;
895                 return 0;
896         case EMULATE_FAIL:
897                 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
898                 break;
899         default:
900                 BUG();
901         }
902
903         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
904         return 0;
905 }
906
907 static int shutdown_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
908 {
909         /*
910          * VMCB is undefined after a SHUTDOWN intercept
911          * so reinitialize it.
912          */
913         memset(vcpu->svm->vmcb, 0, PAGE_SIZE);
914         init_vmcb(vcpu->svm->vmcb);
915
916         kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
917         return 0;
918 }
919
920 static int io_get_override(struct kvm_vcpu *vcpu,
921                           struct vmcb_seg **seg,
922                           int *addr_override)
923 {
924         u8 inst[MAX_INST_SIZE];
925         unsigned ins_length;
926         gva_t rip;
927         int i;
928
929         rip =  vcpu->svm->vmcb->save.rip;
930         ins_length = vcpu->svm->next_rip - rip;
931         rip += vcpu->svm->vmcb->save.cs.base;
932
933         if (ins_length > MAX_INST_SIZE)
934                 printk(KERN_DEBUG
935                        "%s: inst length err, cs base 0x%llx rip 0x%llx "
936                        "next rip 0x%llx ins_length %u\n",
937                        __FUNCTION__,
938                        vcpu->svm->vmcb->save.cs.base,
939                        vcpu->svm->vmcb->save.rip,
940                        vcpu->svm->vmcb->control.exit_info_2,
941                        ins_length);
942
943         if (kvm_read_guest(vcpu, rip, ins_length, inst) != ins_length)
944                 /* #PF */
945                 return 0;
946
947         *addr_override = 0;
948         *seg = NULL;
949         for (i = 0; i < ins_length; i++)
950                 switch (inst[i]) {
951                 case 0xf0:
952                 case 0xf2:
953                 case 0xf3:
954                 case 0x66:
955                         continue;
956                 case 0x67:
957                         *addr_override = 1;
958                         continue;
959                 case 0x2e:
960                         *seg = &vcpu->svm->vmcb->save.cs;
961                         continue;
962                 case 0x36:
963                         *seg = &vcpu->svm->vmcb->save.ss;
964                         continue;
965                 case 0x3e:
966                         *seg = &vcpu->svm->vmcb->save.ds;
967                         continue;
968                 case 0x26:
969                         *seg = &vcpu->svm->vmcb->save.es;
970                         continue;
971                 case 0x64:
972                         *seg = &vcpu->svm->vmcb->save.fs;
973                         continue;
974                 case 0x65:
975                         *seg = &vcpu->svm->vmcb->save.gs;
976                         continue;
977                 default:
978                         return 1;
979                 }
980         printk(KERN_DEBUG "%s: unexpected\n", __FUNCTION__);
981         return 0;
982 }
983
984 static unsigned long io_adress(struct kvm_vcpu *vcpu, int ins, u64 *address)
985 {
986         unsigned long addr_mask;
987         unsigned long *reg;
988         struct vmcb_seg *seg;
989         int addr_override;
990         struct vmcb_save_area *save_area = &vcpu->svm->vmcb->save;
991         u16 cs_attrib = save_area->cs.attrib;
992         unsigned addr_size = get_addr_size(vcpu);
993
994         if (!io_get_override(vcpu, &seg, &addr_override))
995                 return 0;
996
997         if (addr_override)
998                 addr_size = (addr_size == 2) ? 4: (addr_size >> 1);
999
1000         if (ins) {
1001                 reg = &vcpu->regs[VCPU_REGS_RDI];
1002                 seg = &vcpu->svm->vmcb->save.es;
1003         } else {
1004                 reg = &vcpu->regs[VCPU_REGS_RSI];
1005                 seg = (seg) ? seg : &vcpu->svm->vmcb->save.ds;
1006         }
1007
1008         addr_mask = ~0ULL >> (64 - (addr_size * 8));
1009
1010         if ((cs_attrib & SVM_SELECTOR_L_MASK) &&
1011             !(vcpu->svm->vmcb->save.rflags & X86_EFLAGS_VM)) {
1012                 *address = (*reg & addr_mask);
1013                 return addr_mask;
1014         }
1015
1016         if (!(seg->attrib & SVM_SELECTOR_P_SHIFT)) {
1017                 svm_inject_gp(vcpu, 0);
1018                 return 0;
1019         }
1020
1021         *address = (*reg & addr_mask) + seg->base;
1022         return addr_mask;
1023 }
1024
1025 static int io_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1026 {
1027         u32 io_info = vcpu->svm->vmcb->control.exit_info_1; //address size bug?
1028         int _in = io_info & SVM_IOIO_TYPE_MASK;
1029
1030         ++kvm_stat.io_exits;
1031
1032         vcpu->svm->next_rip = vcpu->svm->vmcb->control.exit_info_2;
1033
1034         kvm_run->exit_reason = KVM_EXIT_IO;
1035         kvm_run->io.port = io_info >> 16;
1036         kvm_run->io.direction = (_in) ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
1037         kvm_run->io.size = ((io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT);
1038         kvm_run->io.string = (io_info & SVM_IOIO_STR_MASK) != 0;
1039         kvm_run->io.rep = (io_info & SVM_IOIO_REP_MASK) != 0;
1040
1041         if (kvm_run->io.string) {
1042                 unsigned addr_mask;
1043
1044                 addr_mask = io_adress(vcpu, _in, &kvm_run->io.address);
1045                 if (!addr_mask) {
1046                         printk(KERN_DEBUG "%s: get io address failed\n",
1047                                __FUNCTION__);
1048                         return 1;
1049                 }
1050
1051                 if (kvm_run->io.rep) {
1052                         kvm_run->io.count
1053                                 = vcpu->regs[VCPU_REGS_RCX] & addr_mask;
1054                         kvm_run->io.string_down = (vcpu->svm->vmcb->save.rflags
1055                                                    & X86_EFLAGS_DF) != 0;
1056                 }
1057         } else
1058                 kvm_run->io.value = vcpu->svm->vmcb->save.rax;
1059         return 0;
1060 }
1061
1062 static int nop_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1063 {
1064         return 1;
1065 }
1066
1067 static int halt_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1068 {
1069         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 1;
1070         skip_emulated_instruction(vcpu);
1071         if (vcpu->irq_summary)
1072                 return 1;
1073
1074         kvm_run->exit_reason = KVM_EXIT_HLT;
1075         ++kvm_stat.halt_exits;
1076         return 0;
1077 }
1078
1079 static int invalid_op_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1080 {
1081         inject_ud(vcpu);
1082         return 1;
1083 }
1084
1085 static int task_switch_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1086 {
1087         printk(KERN_DEBUG "%s: task swiche is unsupported\n", __FUNCTION__);
1088         kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1089         return 0;
1090 }
1091
1092 static int cpuid_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1093 {
1094         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1095         kvm_run->exit_reason = KVM_EXIT_CPUID;
1096         return 0;
1097 }
1098
1099 static int emulate_on_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1100 {
1101         if (emulate_instruction(vcpu, NULL, 0, 0) != EMULATE_DONE)
1102                 printk(KERN_ERR "%s: failed\n", __FUNCTION__);
1103         return 1;
1104 }
1105
1106 static int svm_get_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 *data)
1107 {
1108         switch (ecx) {
1109         case MSR_IA32_TIME_STAMP_COUNTER: {
1110                 u64 tsc;
1111
1112                 rdtscll(tsc);
1113                 *data = vcpu->svm->vmcb->control.tsc_offset + tsc;
1114                 break;
1115         }
1116         case MSR_K6_STAR:
1117                 *data = vcpu->svm->vmcb->save.star;
1118                 break;
1119 #ifdef CONFIG_X86_64
1120         case MSR_LSTAR:
1121                 *data = vcpu->svm->vmcb->save.lstar;
1122                 break;
1123         case MSR_CSTAR:
1124                 *data = vcpu->svm->vmcb->save.cstar;
1125                 break;
1126         case MSR_KERNEL_GS_BASE:
1127                 *data = vcpu->svm->vmcb->save.kernel_gs_base;
1128                 break;
1129         case MSR_SYSCALL_MASK:
1130                 *data = vcpu->svm->vmcb->save.sfmask;
1131                 break;
1132 #endif
1133         case MSR_IA32_SYSENTER_CS:
1134                 *data = vcpu->svm->vmcb->save.sysenter_cs;
1135                 break;
1136         case MSR_IA32_SYSENTER_EIP:
1137                 *data = vcpu->svm->vmcb->save.sysenter_eip;
1138                 break;
1139         case MSR_IA32_SYSENTER_ESP:
1140                 *data = vcpu->svm->vmcb->save.sysenter_esp;
1141                 break;
1142         default:
1143                 return kvm_get_msr_common(vcpu, ecx, data);
1144         }
1145         return 0;
1146 }
1147
1148 static int rdmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1149 {
1150         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1151         u64 data;
1152
1153         if (svm_get_msr(vcpu, ecx, &data))
1154                 svm_inject_gp(vcpu, 0);
1155         else {
1156                 vcpu->svm->vmcb->save.rax = data & 0xffffffff;
1157                 vcpu->regs[VCPU_REGS_RDX] = data >> 32;
1158                 vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1159                 skip_emulated_instruction(vcpu);
1160         }
1161         return 1;
1162 }
1163
1164 static int svm_set_msr(struct kvm_vcpu *vcpu, unsigned ecx, u64 data)
1165 {
1166         switch (ecx) {
1167         case MSR_IA32_TIME_STAMP_COUNTER: {
1168                 u64 tsc;
1169
1170                 rdtscll(tsc);
1171                 vcpu->svm->vmcb->control.tsc_offset = data - tsc;
1172                 break;
1173         }
1174         case MSR_K6_STAR:
1175                 vcpu->svm->vmcb->save.star = data;
1176                 break;
1177 #ifdef CONFIG_X86_64
1178         case MSR_LSTAR:
1179                 vcpu->svm->vmcb->save.lstar = data;
1180                 break;
1181         case MSR_CSTAR:
1182                 vcpu->svm->vmcb->save.cstar = data;
1183                 break;
1184         case MSR_KERNEL_GS_BASE:
1185                 vcpu->svm->vmcb->save.kernel_gs_base = data;
1186                 break;
1187         case MSR_SYSCALL_MASK:
1188                 vcpu->svm->vmcb->save.sfmask = data;
1189                 break;
1190 #endif
1191         case MSR_IA32_SYSENTER_CS:
1192                 vcpu->svm->vmcb->save.sysenter_cs = data;
1193                 break;
1194         case MSR_IA32_SYSENTER_EIP:
1195                 vcpu->svm->vmcb->save.sysenter_eip = data;
1196                 break;
1197         case MSR_IA32_SYSENTER_ESP:
1198                 vcpu->svm->vmcb->save.sysenter_esp = data;
1199                 break;
1200         default:
1201                 return kvm_set_msr_common(vcpu, ecx, data);
1202         }
1203         return 0;
1204 }
1205
1206 static int wrmsr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1207 {
1208         u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1209         u64 data = (vcpu->svm->vmcb->save.rax & -1u)
1210                 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1211         vcpu->svm->next_rip = vcpu->svm->vmcb->save.rip + 2;
1212         if (svm_set_msr(vcpu, ecx, data))
1213                 svm_inject_gp(vcpu, 0);
1214         else
1215                 skip_emulated_instruction(vcpu);
1216         return 1;
1217 }
1218
1219 static int msr_interception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1220 {
1221         if (vcpu->svm->vmcb->control.exit_info_1)
1222                 return wrmsr_interception(vcpu, kvm_run);
1223         else
1224                 return rdmsr_interception(vcpu, kvm_run);
1225 }
1226
1227 static int interrupt_window_interception(struct kvm_vcpu *vcpu,
1228                                    struct kvm_run *kvm_run)
1229 {
1230         /*
1231          * If the user space waits to inject interrupts, exit as soon as
1232          * possible
1233          */
1234         if (kvm_run->request_interrupt_window &&
1235             !vcpu->irq_summary) {
1236                 ++kvm_stat.irq_window_exits;
1237                 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1238                 return 0;
1239         }
1240
1241         return 1;
1242 }
1243
1244 static int (*svm_exit_handlers[])(struct kvm_vcpu *vcpu,
1245                                       struct kvm_run *kvm_run) = {
1246         [SVM_EXIT_READ_CR0]                     = emulate_on_interception,
1247         [SVM_EXIT_READ_CR3]                     = emulate_on_interception,
1248         [SVM_EXIT_READ_CR4]                     = emulate_on_interception,
1249         /* for now: */
1250         [SVM_EXIT_WRITE_CR0]                    = emulate_on_interception,
1251         [SVM_EXIT_WRITE_CR3]                    = emulate_on_interception,
1252         [SVM_EXIT_WRITE_CR4]                    = emulate_on_interception,
1253         [SVM_EXIT_READ_DR0]                     = emulate_on_interception,
1254         [SVM_EXIT_READ_DR1]                     = emulate_on_interception,
1255         [SVM_EXIT_READ_DR2]                     = emulate_on_interception,
1256         [SVM_EXIT_READ_DR3]                     = emulate_on_interception,
1257         [SVM_EXIT_WRITE_DR0]                    = emulate_on_interception,
1258         [SVM_EXIT_WRITE_DR1]                    = emulate_on_interception,
1259         [SVM_EXIT_WRITE_DR2]                    = emulate_on_interception,
1260         [SVM_EXIT_WRITE_DR3]                    = emulate_on_interception,
1261         [SVM_EXIT_WRITE_DR5]                    = emulate_on_interception,
1262         [SVM_EXIT_WRITE_DR7]                    = emulate_on_interception,
1263         [SVM_EXIT_EXCP_BASE + PF_VECTOR]        = pf_interception,
1264         [SVM_EXIT_INTR]                         = nop_on_interception,
1265         [SVM_EXIT_NMI]                          = nop_on_interception,
1266         [SVM_EXIT_SMI]                          = nop_on_interception,
1267         [SVM_EXIT_INIT]                         = nop_on_interception,
1268         [SVM_EXIT_VINTR]                        = interrupt_window_interception,
1269         /* [SVM_EXIT_CR0_SEL_WRITE]             = emulate_on_interception, */
1270         [SVM_EXIT_CPUID]                        = cpuid_interception,
1271         [SVM_EXIT_HLT]                          = halt_interception,
1272         [SVM_EXIT_INVLPG]                       = emulate_on_interception,
1273         [SVM_EXIT_INVLPGA]                      = invalid_op_interception,
1274         [SVM_EXIT_IOIO]                         = io_interception,
1275         [SVM_EXIT_MSR]                          = msr_interception,
1276         [SVM_EXIT_TASK_SWITCH]                  = task_switch_interception,
1277         [SVM_EXIT_SHUTDOWN]                     = shutdown_interception,
1278         [SVM_EXIT_VMRUN]                        = invalid_op_interception,
1279         [SVM_EXIT_VMMCALL]                      = invalid_op_interception,
1280         [SVM_EXIT_VMLOAD]                       = invalid_op_interception,
1281         [SVM_EXIT_VMSAVE]                       = invalid_op_interception,
1282         [SVM_EXIT_STGI]                         = invalid_op_interception,
1283         [SVM_EXIT_CLGI]                         = invalid_op_interception,
1284         [SVM_EXIT_SKINIT]                       = invalid_op_interception,
1285 };
1286
1287
1288 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1289 {
1290         u32 exit_code = vcpu->svm->vmcb->control.exit_code;
1291
1292         kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1293
1294         if (is_external_interrupt(vcpu->svm->vmcb->control.exit_int_info) &&
1295             exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR)
1296                 printk(KERN_ERR "%s: unexpected exit_ini_info 0x%x "
1297                        "exit_code 0x%x\n",
1298                        __FUNCTION__, vcpu->svm->vmcb->control.exit_int_info,
1299                        exit_code);
1300
1301         if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
1302             || svm_exit_handlers[exit_code] == 0) {
1303                 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1304                 printk(KERN_ERR "%s: 0x%x @ 0x%llx cr0 0x%lx rflags 0x%llx\n",
1305                        __FUNCTION__,
1306                        exit_code,
1307                        vcpu->svm->vmcb->save.rip,
1308                        vcpu->cr0,
1309                        vcpu->svm->vmcb->save.rflags);
1310                 return 0;
1311         }
1312
1313         return svm_exit_handlers[exit_code](vcpu, kvm_run);
1314 }
1315
1316 static void reload_tss(struct kvm_vcpu *vcpu)
1317 {
1318         int cpu = raw_smp_processor_id();
1319
1320         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1321         svm_data->tss_desc->type = 9; //available 32/64-bit TSS
1322         load_TR_desc();
1323 }
1324
1325 static void pre_svm_run(struct kvm_vcpu *vcpu)
1326 {
1327         int cpu = raw_smp_processor_id();
1328
1329         struct svm_cpu_data *svm_data = per_cpu(svm_data, cpu);
1330
1331         vcpu->svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
1332         if (vcpu->cpu != cpu ||
1333             vcpu->svm->asid_generation != svm_data->asid_generation)
1334                 new_asid(vcpu, svm_data);
1335 }
1336
1337
1338 static inline void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1339 {
1340         struct vmcb_control_area *control;
1341
1342         control = &vcpu->svm->vmcb->control;
1343         control->int_vector = pop_irq(vcpu);
1344         control->int_ctl &= ~V_INTR_PRIO_MASK;
1345         control->int_ctl |= V_IRQ_MASK |
1346                 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1347 }
1348
1349 static void kvm_reput_irq(struct kvm_vcpu *vcpu)
1350 {
1351         struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1352
1353         if (control->int_ctl & V_IRQ_MASK) {
1354                 control->int_ctl &= ~V_IRQ_MASK;
1355                 push_irq(vcpu, control->int_vector);
1356         }
1357
1358         vcpu->interrupt_window_open =
1359                 !(control->int_state & SVM_INTERRUPT_SHADOW_MASK);
1360 }
1361
1362 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1363                                        struct kvm_run *kvm_run)
1364 {
1365         struct vmcb_control_area *control = &vcpu->svm->vmcb->control;
1366
1367         vcpu->interrupt_window_open =
1368                 (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
1369                  (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1370
1371         if (vcpu->interrupt_window_open && vcpu->irq_summary)
1372                 /*
1373                  * If interrupts enabled, and not blocked by sti or mov ss. Good.
1374                  */
1375                 kvm_do_inject_irq(vcpu);
1376
1377         /*
1378          * Interrupts blocked.  Wait for unblock.
1379          */
1380         if (!vcpu->interrupt_window_open &&
1381             (vcpu->irq_summary || kvm_run->request_interrupt_window)) {
1382                 control->intercept |= 1ULL << INTERCEPT_VINTR;
1383         } else
1384                 control->intercept &= ~(1ULL << INTERCEPT_VINTR);
1385 }
1386
1387 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1388                               struct kvm_run *kvm_run)
1389 {
1390         kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1391                                                   vcpu->irq_summary == 0);
1392         kvm_run->if_flag = (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF) != 0;
1393         kvm_run->cr8 = vcpu->cr8;
1394         kvm_run->apic_base = vcpu->apic_base;
1395 }
1396
1397 /*
1398  * Check if userspace requested an interrupt window, and that the
1399  * interrupt window is open.
1400  *
1401  * No need to exit to userspace if we already have an interrupt queued.
1402  */
1403 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1404                                           struct kvm_run *kvm_run)
1405 {
1406         return (!vcpu->irq_summary &&
1407                 kvm_run->request_interrupt_window &&
1408                 vcpu->interrupt_window_open &&
1409                 (vcpu->svm->vmcb->save.rflags & X86_EFLAGS_IF));
1410 }
1411
1412 static void save_db_regs(unsigned long *db_regs)
1413 {
1414         asm volatile ("mov %%dr0, %0" : "=r"(db_regs[0]));
1415         asm volatile ("mov %%dr1, %0" : "=r"(db_regs[1]));
1416         asm volatile ("mov %%dr2, %0" : "=r"(db_regs[2]));
1417         asm volatile ("mov %%dr3, %0" : "=r"(db_regs[3]));
1418 }
1419
1420 static void load_db_regs(unsigned long *db_regs)
1421 {
1422         asm volatile ("mov %0, %%dr0" : : "r"(db_regs[0]));
1423         asm volatile ("mov %0, %%dr1" : : "r"(db_regs[1]));
1424         asm volatile ("mov %0, %%dr2" : : "r"(db_regs[2]));
1425         asm volatile ("mov %0, %%dr3" : : "r"(db_regs[3]));
1426 }
1427
1428 static int svm_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1429 {
1430         u16 fs_selector;
1431         u16 gs_selector;
1432         u16 ldt_selector;
1433         int r;
1434
1435 again:
1436         if (!vcpu->mmio_read_completed)
1437                 do_interrupt_requests(vcpu, kvm_run);
1438
1439         clgi();
1440
1441         pre_svm_run(vcpu);
1442
1443         save_host_msrs(vcpu);
1444         fs_selector = read_fs();
1445         gs_selector = read_gs();
1446         ldt_selector = read_ldt();
1447         vcpu->svm->host_cr2 = kvm_read_cr2();
1448         vcpu->svm->host_dr6 = read_dr6();
1449         vcpu->svm->host_dr7 = read_dr7();
1450         vcpu->svm->vmcb->save.cr2 = vcpu->cr2;
1451
1452         if (vcpu->svm->vmcb->save.dr7 & 0xff) {
1453                 write_dr7(0);
1454                 save_db_regs(vcpu->svm->host_db_regs);
1455                 load_db_regs(vcpu->svm->db_regs);
1456         }
1457
1458         fx_save(vcpu->host_fx_image);
1459         fx_restore(vcpu->guest_fx_image);
1460
1461         asm volatile (
1462 #ifdef CONFIG_X86_64
1463                 "push %%rbx; push %%rcx; push %%rdx;"
1464                 "push %%rsi; push %%rdi; push %%rbp;"
1465                 "push %%r8;  push %%r9;  push %%r10; push %%r11;"
1466                 "push %%r12; push %%r13; push %%r14; push %%r15;"
1467 #else
1468                 "push %%ebx; push %%ecx; push %%edx;"
1469                 "push %%esi; push %%edi; push %%ebp;"
1470 #endif
1471
1472 #ifdef CONFIG_X86_64
1473                 "mov %c[rbx](%[vcpu]), %%rbx \n\t"
1474                 "mov %c[rcx](%[vcpu]), %%rcx \n\t"
1475                 "mov %c[rdx](%[vcpu]), %%rdx \n\t"
1476                 "mov %c[rsi](%[vcpu]), %%rsi \n\t"
1477                 "mov %c[rdi](%[vcpu]), %%rdi \n\t"
1478                 "mov %c[rbp](%[vcpu]), %%rbp \n\t"
1479                 "mov %c[r8](%[vcpu]),  %%r8  \n\t"
1480                 "mov %c[r9](%[vcpu]),  %%r9  \n\t"
1481                 "mov %c[r10](%[vcpu]), %%r10 \n\t"
1482                 "mov %c[r11](%[vcpu]), %%r11 \n\t"
1483                 "mov %c[r12](%[vcpu]), %%r12 \n\t"
1484                 "mov %c[r13](%[vcpu]), %%r13 \n\t"
1485                 "mov %c[r14](%[vcpu]), %%r14 \n\t"
1486                 "mov %c[r15](%[vcpu]), %%r15 \n\t"
1487 #else
1488                 "mov %c[rbx](%[vcpu]), %%ebx \n\t"
1489                 "mov %c[rcx](%[vcpu]), %%ecx \n\t"
1490                 "mov %c[rdx](%[vcpu]), %%edx \n\t"
1491                 "mov %c[rsi](%[vcpu]), %%esi \n\t"
1492                 "mov %c[rdi](%[vcpu]), %%edi \n\t"
1493                 "mov %c[rbp](%[vcpu]), %%ebp \n\t"
1494 #endif
1495
1496 #ifdef CONFIG_X86_64
1497                 /* Enter guest mode */
1498                 "push %%rax \n\t"
1499                 "mov %c[svm](%[vcpu]), %%rax \n\t"
1500                 "mov %c[vmcb](%%rax), %%rax \n\t"
1501                 SVM_VMLOAD "\n\t"
1502                 SVM_VMRUN "\n\t"
1503                 SVM_VMSAVE "\n\t"
1504                 "pop %%rax \n\t"
1505 #else
1506                 /* Enter guest mode */
1507                 "push %%eax \n\t"
1508                 "mov %c[svm](%[vcpu]), %%eax \n\t"
1509                 "mov %c[vmcb](%%eax), %%eax \n\t"
1510                 SVM_VMLOAD "\n\t"
1511                 SVM_VMRUN "\n\t"
1512                 SVM_VMSAVE "\n\t"
1513                 "pop %%eax \n\t"
1514 #endif
1515
1516                 /* Save guest registers, load host registers */
1517 #ifdef CONFIG_X86_64
1518                 "mov %%rbx, %c[rbx](%[vcpu]) \n\t"
1519                 "mov %%rcx, %c[rcx](%[vcpu]) \n\t"
1520                 "mov %%rdx, %c[rdx](%[vcpu]) \n\t"
1521                 "mov %%rsi, %c[rsi](%[vcpu]) \n\t"
1522                 "mov %%rdi, %c[rdi](%[vcpu]) \n\t"
1523                 "mov %%rbp, %c[rbp](%[vcpu]) \n\t"
1524                 "mov %%r8,  %c[r8](%[vcpu]) \n\t"
1525                 "mov %%r9,  %c[r9](%[vcpu]) \n\t"
1526                 "mov %%r10, %c[r10](%[vcpu]) \n\t"
1527                 "mov %%r11, %c[r11](%[vcpu]) \n\t"
1528                 "mov %%r12, %c[r12](%[vcpu]) \n\t"
1529                 "mov %%r13, %c[r13](%[vcpu]) \n\t"
1530                 "mov %%r14, %c[r14](%[vcpu]) \n\t"
1531                 "mov %%r15, %c[r15](%[vcpu]) \n\t"
1532
1533                 "pop  %%r15; pop  %%r14; pop  %%r13; pop  %%r12;"
1534                 "pop  %%r11; pop  %%r10; pop  %%r9;  pop  %%r8;"
1535                 "pop  %%rbp; pop  %%rdi; pop  %%rsi;"
1536                 "pop  %%rdx; pop  %%rcx; pop  %%rbx; \n\t"
1537 #else
1538                 "mov %%ebx, %c[rbx](%[vcpu]) \n\t"
1539                 "mov %%ecx, %c[rcx](%[vcpu]) \n\t"
1540                 "mov %%edx, %c[rdx](%[vcpu]) \n\t"
1541                 "mov %%esi, %c[rsi](%[vcpu]) \n\t"
1542                 "mov %%edi, %c[rdi](%[vcpu]) \n\t"
1543                 "mov %%ebp, %c[rbp](%[vcpu]) \n\t"
1544
1545                 "pop  %%ebp; pop  %%edi; pop  %%esi;"
1546                 "pop  %%edx; pop  %%ecx; pop  %%ebx; \n\t"
1547 #endif
1548                 :
1549                 : [vcpu]"a"(vcpu),
1550                   [svm]"i"(offsetof(struct kvm_vcpu, svm)),
1551                   [vmcb]"i"(offsetof(struct vcpu_svm, vmcb_pa)),
1552                   [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1553                   [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1554                   [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1555                   [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1556                   [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1557                   [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP]))
1558 #ifdef CONFIG_X86_64
1559                   ,[r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1560                   [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1561                   [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1562                   [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1563                   [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1564                   [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1565                   [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1566                   [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15]))
1567 #endif
1568                 : "cc", "memory" );
1569
1570         fx_save(vcpu->guest_fx_image);
1571         fx_restore(vcpu->host_fx_image);
1572
1573         if ((vcpu->svm->vmcb->save.dr7 & 0xff))
1574                 load_db_regs(vcpu->svm->host_db_regs);
1575
1576         vcpu->cr2 = vcpu->svm->vmcb->save.cr2;
1577
1578         write_dr6(vcpu->svm->host_dr6);
1579         write_dr7(vcpu->svm->host_dr7);
1580         kvm_write_cr2(vcpu->svm->host_cr2);
1581
1582         load_fs(fs_selector);
1583         load_gs(gs_selector);
1584         load_ldt(ldt_selector);
1585         load_host_msrs(vcpu);
1586
1587         reload_tss(vcpu);
1588
1589         /*
1590          * Profile KVM exit RIPs:
1591          */
1592         if (unlikely(prof_on == KVM_PROFILING))
1593                 profile_hit(KVM_PROFILING,
1594                         (void *)(unsigned long)vcpu->svm->vmcb->save.rip);
1595
1596         stgi();
1597
1598         kvm_reput_irq(vcpu);
1599
1600         vcpu->svm->next_rip = 0;
1601
1602         if (vcpu->svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
1603                 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1604                 kvm_run->exit_reason = vcpu->svm->vmcb->control.exit_code;
1605                 post_kvm_run_save(vcpu, kvm_run);
1606                 return 0;
1607         }
1608
1609         r = handle_exit(vcpu, kvm_run);
1610         if (r > 0) {
1611                 if (signal_pending(current)) {
1612                         ++kvm_stat.signal_exits;
1613                         post_kvm_run_save(vcpu, kvm_run);
1614                         return -EINTR;
1615                 }
1616
1617                 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
1618                         ++kvm_stat.request_irq_exits;
1619                         post_kvm_run_save(vcpu, kvm_run);
1620                         return -EINTR;
1621                 }
1622                 kvm_resched(vcpu);
1623                 goto again;
1624         }
1625         post_kvm_run_save(vcpu, kvm_run);
1626         return r;
1627 }
1628
1629 static void svm_flush_tlb(struct kvm_vcpu *vcpu)
1630 {
1631         force_new_asid(vcpu);
1632 }
1633
1634 static void svm_set_cr3(struct kvm_vcpu *vcpu, unsigned long root)
1635 {
1636         vcpu->svm->vmcb->save.cr3 = root;
1637         force_new_asid(vcpu);
1638 }
1639
1640 static void svm_inject_page_fault(struct kvm_vcpu *vcpu,
1641                                   unsigned long  addr,
1642                                   uint32_t err_code)
1643 {
1644         uint32_t exit_int_info = vcpu->svm->vmcb->control.exit_int_info;
1645
1646         ++kvm_stat.pf_guest;
1647
1648         if (is_page_fault(exit_int_info)) {
1649
1650                 vcpu->svm->vmcb->control.event_inj_err = 0;
1651                 vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1652                                                         SVM_EVTINJ_VALID_ERR |
1653                                                         SVM_EVTINJ_TYPE_EXEPT |
1654                                                         DF_VECTOR;
1655                 return;
1656         }
1657         vcpu->cr2 = addr;
1658         vcpu->svm->vmcb->save.cr2 = addr;
1659         vcpu->svm->vmcb->control.event_inj =    SVM_EVTINJ_VALID |
1660                                                 SVM_EVTINJ_VALID_ERR |
1661                                                 SVM_EVTINJ_TYPE_EXEPT |
1662                                                 PF_VECTOR;
1663         vcpu->svm->vmcb->control.event_inj_err = err_code;
1664 }
1665
1666
1667 static int is_disabled(void)
1668 {
1669         return 0;
1670 }
1671
1672 static void
1673 svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1674 {
1675         /*
1676          * Patch in the VMMCALL instruction:
1677          */
1678         hypercall[0] = 0x0f;
1679         hypercall[1] = 0x01;
1680         hypercall[2] = 0xd9;
1681         hypercall[3] = 0xc3;
1682 }
1683
1684 static struct kvm_arch_ops svm_arch_ops = {
1685         .cpu_has_kvm_support = has_svm,
1686         .disabled_by_bios = is_disabled,
1687         .hardware_setup = svm_hardware_setup,
1688         .hardware_unsetup = svm_hardware_unsetup,
1689         .hardware_enable = svm_hardware_enable,
1690         .hardware_disable = svm_hardware_disable,
1691
1692         .vcpu_create = svm_create_vcpu,
1693         .vcpu_free = svm_free_vcpu,
1694
1695         .vcpu_load = svm_vcpu_load,
1696         .vcpu_put = svm_vcpu_put,
1697         .vcpu_decache = svm_vcpu_decache,
1698
1699         .set_guest_debug = svm_guest_debug,
1700         .get_msr = svm_get_msr,
1701         .set_msr = svm_set_msr,
1702         .get_segment_base = svm_get_segment_base,
1703         .get_segment = svm_get_segment,
1704         .set_segment = svm_set_segment,
1705         .get_cs_db_l_bits = svm_get_cs_db_l_bits,
1706         .decache_cr0_cr4_guest_bits = svm_decache_cr0_cr4_guest_bits,
1707         .set_cr0 = svm_set_cr0,
1708         .set_cr0_no_modeswitch = svm_set_cr0,
1709         .set_cr3 = svm_set_cr3,
1710         .set_cr4 = svm_set_cr4,
1711         .set_efer = svm_set_efer,
1712         .get_idt = svm_get_idt,
1713         .set_idt = svm_set_idt,
1714         .get_gdt = svm_get_gdt,
1715         .set_gdt = svm_set_gdt,
1716         .get_dr = svm_get_dr,
1717         .set_dr = svm_set_dr,
1718         .cache_regs = svm_cache_regs,
1719         .decache_regs = svm_decache_regs,
1720         .get_rflags = svm_get_rflags,
1721         .set_rflags = svm_set_rflags,
1722
1723         .invlpg = svm_invlpg,
1724         .tlb_flush = svm_flush_tlb,
1725         .inject_page_fault = svm_inject_page_fault,
1726
1727         .inject_gp = svm_inject_gp,
1728
1729         .run = svm_vcpu_run,
1730         .skip_emulated_instruction = skip_emulated_instruction,
1731         .vcpu_setup = svm_vcpu_setup,
1732         .patch_hypercall = svm_patch_hypercall,
1733 };
1734
1735 static int __init svm_init(void)
1736 {
1737         return kvm_init_arch(&svm_arch_ops, THIS_MODULE);
1738 }
1739
1740 static void __exit svm_exit(void)
1741 {
1742         kvm_exit_arch();
1743 }
1744
1745 module_init(svm_init)
1746 module_exit(svm_exit)